CJ ONSTYLE Quadrupled Its AI Traffic by Rewriting the Catalogue
CJ ONSTYLE's traffic from conversational AI platforms rose more than fourfold between January and late May, and it shipped a dedicated ChatGPT app on the fifteenth. The engine was 600,000 rewritten product descriptions, not a new model: a thirty-year home-shopping incumbent reaching conversational commerce by doing the clerical work many Western retailers keep deferring.
Sir John Crabstone
CJ ONSTYLE has done what most Western retailers only describe: it shipped a store inside ChatGPT and then published the lift. Traffic from conversational AI platforms ran more than four times higher over the first twenty-five days of May than across the same dates in January. The dedicated app went live on the fifteenth. The multiple matters less than the baseline behind it: a four-month before-and-after that most retailers decline to publish.
The inflow came from more than one assistant: ChatGPT and Gemini both. CJ optimised its catalogue for any model a shopper might open, then launched its dedicated app on the largest. It approached the assistants the way it once approached search: optimise for all of them, commit to none.
The fourfold is the headline. The number that explains it is six hundred thousand: the products CJ has rewritten for the ChatGPT app, on the way to a million by year-end. The rewriting was clerical. Product copy was recast into the phrases shoppers type, like “office slacks” and “summer wedding guest look”. From the reviews, CJ pulled what a spec sheet omits: “low noise”, “less skin irritation”. That is not a model breakthrough — it is copy-editing at industrial scale.
CJ’s own app was already compounding before the assistants arrived. Its in-app recommendation area lifted sales 37% year on year, off the same investment in customer data. Clean product data serves the recommender within and the assistant without. Two engines now run on one catalogue.
Conversational shopping rewards whoever owns the most product language and the least vanity about rewriting it; a home-shopping channel has decades of one and no use for the other.
By the last decade’s story, the model-rich startup was supposed to win this. Conversational commerce instead favours whoever holds the deepest catalogue, the incumbent every disruption memo wrote off. The vendors selling models had an interest in the model being the hard part. We have argued in these pages that the bottleneck moved from the model to the catalogue; CJ ONSTYLE is the first incumbent to act as though it agreed.
The West has been celebrating a smaller number. A fifth of Walmart’s referral clicks came from ChatGPT last August, the figure the trade press kept quoting. Those clicks were under five percent of all visits. CJ did not wait to be found; it built the surface that does the finding.
Korean retail is moving in a cluster. LF Mall and Lotte Home Shopping both launched dedicated ChatGPT apps in May; Shinsegae’s OpenAI tie-up still points at full-scale ChatGPT commerce in 2027, with E-Mart’s shopping agent only promised for later this year. The format built to sell by talking needed one new habit: to write down what it had spent thirty years saying aloud.